I have the same cards...
I've observed only about 2 frames of delay - from input to output...
This of course is going to be affected by everything you have in line:
1 - 2 frames right from the camera (really test it!)
1 frame from the switcher
If you have any scalers or converters inline (before or after your server)
1 frame each.
Fluid frame also seems to add another frame.
Then finally the card with 2 frames.
If you add all of these up that's about 8 frames if you've got a scaler in line and are assuming 2 frames from the camera (which is usually the case).
One thing that we've noticed is that de-interlacing also seems to add a frame.
What we have done to combat this is to change everything in the camera and switching world to 720p.
At first this seemed quite counter intuitive, we want it to look good, however we did notice that we're always scaling our capture to a smaller size anyways - it is not ever 1920 x 1080 on screen, it is in fact closer to 1280 x 720 or smaller in their respective "PIPs".
If you take a look at this post:
http://forum.coolux.de/viewtopic.php?f=4&t=709&p=2223&hilit=sdi+decklink+duo#p2223 you'll see that I've been getting 2 frames for quite a while - and that is on a now 5 year old machine.
If you set a benchmark like this - shoot timecode with your source camera, put it on screen and take a fast picture or high speed video recording with both of them in shot - you can see where your delay comes from and how what you're doing is either improving it or making it worse.
I hope this helps,